The Energy Content and Composition of Meals Consumed after an Overnight Fast and Their Effects on Diet Induced Thermogenesis: A Systematic Review, Meta-Analyses and Meta-Regressions

This systematic review investigated the effects of differing energy intakes, macronutrient compositions, and eating patterns of meals consumed after an overnight fast on Diet Induced Thermogenesis (DIT). The initial search identified 2482 records; 26 papers remained once duplicates were removed and inclusion criteria were applied. Studies (n = 27) in the analyses were randomized crossover designs comparing the effects of two or more eating events on DIT. Higher energy intake increased DIT; in a mixed model meta-regression, for every 100 kJ increase in energy intake, DIT increased by 1.1 kJ/h (p < 0.001). Meals with a high protein or carbohydrate content had a higher DIT than high fat, although this effect was not always significant. Meals with medium chain triglycerides had a significantly higher DIT than long chain triglycerides (meta-analysis, p = 0.002). Consuming the same meal as a single bolus eating event compared to multiple small meals or snacks was associated with a significantly higher DIT (meta-analysis, p = 0.02). Unclear or inconsistent findings were found by comparing the consumption of meals quickly or slowly, and palatability was not significantly associated with DIT. These findings indicate that the magnitude of the increase in DIT is influenced by the energy intake, macronutrient composition, and eating pattern of the meal.


Introduction
The meal consumed after an overnight fast, generally referred to as breakfast, is often described as 'the most important meal of the day' [1] as it is believed to contribute to good health and nutrition by providing essential nutrients early in the day [2]. Skipping breakfast is associated with increased weight gain and obesity, suggesting that breakfast may be protective against weight gain [1,3]. Among the explanations for this protective effect of breakfast are that it stimulates the body's metabolism because it breaks the overnight fast [4], potentially contributing to increased total daily energy expenditure. The extent of this effect would depend on the diet induced thermogenesis (DIT) response to the meal consumed. Evidence supporting this proposal is limited and contradictory [5,6]. Alternatively, eating breakfast may result in decreased energy consumption during the rest of the day, however the evidence available from previous trials in this area is also limited and contradictory [1]. use the terms morning meal or breakfast in the article. Also the keywords 'thermic' and 'thermogenic' were added to the search to ensure studies using slightly different language were not missed.

Inclusion and Exclusion Criteria
For this review, only studies designated as level A evidence (randomized controlled trials (RCTs) and randomized crossover trials), as defined by the Academy of Nutrition and Dietetics, and with two or more eating events for comparison, were included. Studies were included if they provided a snack or a meal in the morning after participants fasted overnight. Studies were excluded if they provided infusions, injections, or capsules with the meal (e.g., saline or drug infusion, drug or placebo capsules). Interventions consisting of meals administered as enteral or intranasal or intra-gastric infusion or consisting of supplements instead of meals (e.g., protein or fat or sugar emulsions) or of meals supplemented with other components (e.g., addition of cellulose or pectin) including stimulants (e.g., caffeine, green tea, chilli, capsaicin, alcohol) were excluded. In these studies, the control meal (e.g., oral feeding or meal without stimulants) data were extracted if provided. When studies provided additional non-dietary interventions (e.g., exercise, sleep interventions), only data from the first meal consumed after an overnight fast intervention of the controlled arm were extracted.
Studies were included if they were published in English with male and/or female adult (≥18 years old) human participants. Data from populations such as children, adolescents, athletes or exercise-trained groups, patients with chronic or acute disease, obese individuals, pregnant or lactating women, or smokers were excluded. Studies with mixed populations of healthy weight, overweight, and obese populations were included; however, studies targeting only obese subjects or a mix of overweight and/or obese participants were excluded. Studies were also excluded when the majority of the participants were obese leading to a mean or median body mass index (BMI) ≥30 kg/m 2 . When studies compared specific populations (e.g., obese, pregnant women, athletes, smokers) to a control group, only data from the control group were included. The original search included all study designs, however, only RCT or randomized cross over designs were included in the analyses for this paper. Articles were excluded if they were expert opinion papers or if they described animal, in vitro, or in vivo experimental studies.

Outcome Measures (Dependent Variables)
Diet induced thermogenesis measured by indirect calorimetry was the main outcome measure. Other outcomes of interest were indirect calorimetry fasting RMR and postprandial energy expenditure.

First Meal Consumed after an Overnight Fast (Independent Variables)
The intervention was the first snack or meal of the day consumed in the morning after an overnight fast. Macronutrient compositions were described as percentages of the energy content of the meal. Energy was expressed in kJ.

Systematic Review Process
Titles and abstracts were assessed for full text retrieval (A.Q.). Full text articles were assessed against the inclusion and exclusion criteria by two independent reviewers (A.Q. and A.P.). The quality criteria checklist for primary research of the Academy of Nutrition and Dietetics was used to assess the quality of the included studies by two independent reviewers (A.Q. and L.M.-W. or A.P.). The quality criteria tool assessed the studies for relevance and validity of the selected publications. A study was deemed positive if it met all the priority criteria, at least one of the validity criteria, and all of the relevance questions. A neutral rating indicated that most of the validity criteria were met but the study may not have met one or more of the priority criteria and/or one or more of the relevance questions. A study was rated as negative if six or more of the validity and/or priority criteria were rated negative. Any discrepancies between reviewers at the full text and quality stage were assessed by a fourth reviewer (R.C.) until a consensus decision was reached.

Data Extraction
The relevant data from the studies were extracted into tables (A.Q.) and evaluated for completeness (A.P., R.C., L.M.-W.). The following information was extracted: study design, significance, inclusion and exclusion criteria, country location, sample size, participant characteristics (intervention and comparator groups), recruitment, blinding used, intervention, statistical analysis, timing of measurements, dependent and independent variables, co-variates, length of follow up and results (key findings and other findings), and author conclusions.

Participant Characteristics
Participant characteristics (age, gender, BMI, fat mass (FM), and fat free mass (FFM)) were extracted when provided or calculated from the data provided. BMI was calculated using the WHO criteria as illustrated in Supplementary Materials Table S1. FM and FFM were expressed as % of total body weight or in kg. If only individual participant data were provided, the mean, standard deviation (SD), and standard error (SE) were calculated with the formula described in Supplementary Materials Table S1. The percentage of males in the sample was calculated (100% indicated that only males were recruited and 0% only females).

Characteristics of the Meals
The energy content of the meals was expressed in kJ. The conversion factor of 4.184 was used to convert kcal to kJ. When studies provided the macronutrient composition of meals only in grams, it was converted from g to % of energy using the two formulas described in Supplementary Materials Table S1.

Outcome Characteristics
RMR, also known as Resting Energy Expenditure (REE), is defined as the quantity of energy used to maintain physiological function under resting conditions. DIT, also called the thermic effect of food, postprandial energy expenditure above baseline, or meal-induced thermogenesis, is defined as the increase in RMR as a result of the consumption of food or a meal [12,13]. DIT data were extracted in kJ and/or as the percentage of energy content of the meal (ECM) and/or as the percentage increase above baseline (AB) RMR. When the studies provided only the total postprandial energy expenditure and the RMR, the mean DIT was obtained from the difference of the total postprandial energy expenditure in kJ and the RMR in kJ for the same measure of time. DIT expressed in kJ or as percentage of energy content was divided by the number of hours that DIT was measured, or multiplied by 60 if it was provided in kJ/minute, in order to provide values as kJ per hour or percentage per hour. The formulas described in Supplementary Materials Table S1 were used to convert DIT from one unit of measurement to another when not provided by the authors.

Meta-Regressions
The main outcome variable used in the meta-regression was DIT in kJ/h. Mixed model meta-regression was used to investigate the relationship between energy intake (kJ) after an overnight fast and DIT (kJ/h). The first model conducted was a univariate analysis, which only included DIT (kJ/h) and kJ intake. The second model also included four confounding factors (percentage of males, age, BMI and hours of DIT measurement). These meta-regression models were conducted using Stata/IC 13.1 (StataCorp LP, College Station, TX, USA) and with consultant statistical support.

Meta-Analyses
Fixed model meta-analyses were conducted in Review Manager (RevMan) to determine the mean difference in DIT (kJ/h) of pairs of comparisons. These meta-analyses were conducted with consultant statistical support.

Results
A total of 2482 papers were identified from the four databases searched; 1756 papers remained after duplicates were removed and 351 full text articles were reviewed ( Figure 1). Only 27 Level A evidence studies from 26 papers (one paper described two studies [14]) were relevant to answer the review questions for this paper. Of the 26 papers, four were rated positive [15][16][17][18], none were rated negative, and the remaining 22 papers were rated neutral. Table 1 summarizes the 27 studies for participants' characteristics and study protocols. Table 2 summarizes the interventions and outcomes of the studies. Nine studies were conducted in the USA [19][20][21][22][23][24][25][26][27], five in Japan [14,[28][29][30], four in the UK [17,[31][32][33], two in Australia [18,34], two in France [35,36], two in Denmark [15,37], one in Germany [38], one in Spain [39], and one in the Netherlands [16].The majority of the studies were not blinded, three were double blinded [14,37], and two studies were single blinded [18,24]. One study provided intervention meals for two weeks for each arm [36] whereas all other studies provided only one day interventions.

Comparison and Meta-Regression of the Effects of Higher and Lower Energy Intakes on DIT
Five studies [21][22][23]31,36] with the primary aim of comparing the effects of meals with different energy intakes on DIT were identified. Three studies [23,31,36] found an increased DIT when a higher energy intake was consumed, although only one indicated statistical significance [31]. Kinabo and Durbin [31] compared high CHO, low fat meals at two energy intake levels: 2520 kJ and 5040 kJ, and low CHO, high fat meals at the same two energy intake levels. This study found that a higher energy intake was associated with a significantly (p < 0.001) higher DIT, regardless of dietary composition. Higher energy intake (5040 kJ) resulted in a similar DIT for the high CHO, low fat (71.2 (15.5) kJ/h) and low CHO, high fat (68 (12.4) kJ/h) meals, and this was higher than the DIT for the lower energy intake (2520 kJ) high CHO, low fat (45.6 (9.3) kJ/h) and low CHO, high fat (45.6 (10.8) kJ/h) meals [31].
Martin et al. [36] compared two weeks of low energy, moderate fat meals (418 kJ) to two weeks of high energy, low fat meals (2929 kJ) and found a higher DIT after the high energy, low fat meals (low energy, moderate fat meals 4.5 (1.4) kJ/h; high energy, low fat meals 35.6 (2.6) kJ/h; no p value provided [36]).
Bennet et al. [21] compared a high fat meal (kJ not provided) to a normal fat meal (kJ not provided). The high fat meal was 1881 kJ higher due to the addition of 50 g of fat compared to the normal fat meal. This study did not find any significant differences in DIT (high fat meal 1.2 (0.40) %/h, normal fat meal 1.3 (0.3) %/h; p > 0.05 for %/6 h ECM for all participants, including some trained individuals, for the statistical tests) [21].
Segal et al. [22] compared consuming a meal with a fixed energy intake (3013 kJ) to a meal providing 35% of each individual's 24 h RMR (caloric intake varying between participants, on average 2889 kJ intake) [22]. This study did not find any significant difference in DIT (fixed: 96.3 (17.6) kJ/h; 35% RMR: 89.3 (17.6) kJ/h, p > 0.05 for %/3 h ECM) but there was little difference in the energy intakes between the two meals [22].
In order to further resolve the effect of energy intake on DIT, mixed model meta-regression analyses were undertaken to investigate more broadly the relationship between energy intake (kJ) after an overnight fast and DIT (kJ/h). Two models were produced: the first one included only energy intake (kJ) and the outcome variable DIT (kJ/h); the second model also included four confounding factors (percentage of males, BMI, age, and hours of DIT measurement). Figure 2 represents Model 1 (coefficient 0.011, standard error 0.0013, p < 0.001, 95% confidence interval, (CI) 0.0083; 0.014) conducted for 19 studies [14,[16][17][18][19]22,24,25,27,[30][31][32][33][34][35][36][37][38] with a total of 54 treatment arms. Eight studies could not be included in the meta-analyses because they had missing values for one or more of the variables investigated in the model. This model shows that DIT (kJ) increases significantly (p < 0.001) when the kJ content of meals increases, although this increase is of a small magnitude (coefficient 0.011). This model predicts that for every 100 kJ increase in energy intake, DIT increases by 1.1 kJ/h. Model 2, adjusted for percentage of males, BMI, age, and hours of DIT measurement, also predicted a small but significant increase in DIT for every kJ intake (coefficient 0.012, standard error 0.0013, p < 0.001; CI: 0.0091; 0.014). This model predicts that for every 100 kJ increase in energy intake, DIT increases by 1.2 kJ/h. In this model, 16 studies were included with a total of 48 arms. Three studies included in model 1 were not included in model 2 because they had missing values for one or more of the variables investigated [16,25,34]. DIT accounted for 47.4% of the variance in Model 1 and 70.6% of the variance in Model 2. representing the outcome (DIT); each circle represents the value of DIT (kJ/h) for an arm of a study, and the size of the circle is inversely proportional to the standard error (SE) of the study. The influence of each study on the model depends on the size of the SE. Specifically, a study arm with a large SE is represented in the figure by a small circle, which means that this study arm had a small influence on the model whereas a study arm with a small SE is represented by a large circle, which means that this study arm had a large influence on the model.

Influence of Macronutrient Composition on DIT
Six studies [15,20,25,30,31,33] compared meals differing in macronutrient composition (fat vs. CHO and/or vs. protein). Five of these papers compared consuming a meal high in CHO with a meal high in fat. Nagai et al. [30] reported a higher DIT with a high CHO meal (3255 (306.5) kJ) compared to an isocaloric meal high in fat (3255 (306.5) kJ). DIT was 43.1 (13.7) kJ/h for the high CHO meal and 32.6 (14.1) kJ/h for the high fat meal, p < 0.05 for %/3.5 h ECM [30]. Blundell et al. [33] provided isocaloric comparisons (both meals contained 2092 kJ) and found a statistically significant effect on DIT (high CHO milkshake: habitually high fat consumers 38.2 (26.0) kJ/h and habitually low fat consumers 35.2 (15.6) kJ/h; high fat milkshake: habitually high fat consumers 27.5 (28.9) kJ/h and habitually low fat consumers 25.6 (14.5) kJ/h, p < 0.05 for kJ/day) [33]. The other two studies provided meals with only small differences in energy content (high CHO meal 2068 kJ and high fat meal 2093 kJ) [20]; high CHO meal 3021 (1194.0) kJ and moderate fat meal 2996 (1167.4) kJ) [25]), and DIT was as follows: high CHO meal 54.6 kJ/h, high fat meal 27.8 kJ/h [20]; high CHO meal 57.8 (19.1) kJ/h and moderate fat meal 49.8 (21.6) kJ/h [25]. No p values were provided for these comparisons; therefore, it is not known if these comparisons were statistically significantly different [20,25].
One study provided isocaloric comparisons and found no significant effect on DIT between high CHO, low fat meals and low CHO, high fat meals [31]. This study [31], which was described in Section 3.4 (higher energy vs. lower energy intake), provided the same group of subjects with high CHO, low fat meals of two different energy contents (2510 kJ and 5040 kJ), as well as low CHO, high fat meals of two different energy contents (2520 kJ and 5040 kJ) [31]. The DIT data were as follows: 5040 kJ high CHO, low fat meal 71.2 (15.5) kJ/h and 2520 kJ high CHO, low fat meal 45.6 (9.3) kJ/h vs.  ) representing the outcome (DIT); each circle represents the value of DIT (kJ/h) for an arm of a study, and the size of the circle is inversely proportional to the standard error (SE) of the study. The influence of each study on the model depends on the size of the SE. Specifically, a study arm with a large SE is represented in the figure by a small circle, which means that this study arm had a small influence on the model whereas a study arm with a small SE is represented by a large circle, which means that this study arm had a large influence on the model.

Influence of Macronutrient Composition on DIT
Six studies [15,20,25,30,31,33] compared meals differing in macronutrient composition (fat vs. CHO and/or vs. protein). Five of these papers compared consuming a meal high in CHO with a meal high in fat. Nagai et al. [30] reported a higher DIT with a high CHO meal (3255 (306.5) kJ) compared to an isocaloric meal high in fat (3255 (306.5) kJ). DIT was 43.1 (13.7) kJ/h for the high CHO meal and 32.6 (14.1) kJ/h for the high fat meal, p < 0.05 for %/3.5 h ECM [30]. Blundell et al. [33] provided isocaloric comparisons (both meals contained 2092 kJ) and found a statistically significant effect on DIT (high CHO milkshake: habitually high fat consumers 38.2 (26.0) kJ/h and habitually low fat consumers 35.2 (15.6) kJ/h; high fat milkshake: habitually high fat consumers 27.5 (28.9) kJ/h and habitually low fat consumers 25.6 (14.5) kJ/h, p < 0.05 for kJ/day) [33]. The other two studies provided meals with only small differences in energy content (high CHO meal 2068 kJ and high fat meal 2093 kJ) [20]; high CHO meal 3021 (1194.0) kJ and moderate fat meal 2996 (1167.4) kJ) [25]), and DIT was as follows: high CHO meal 54.6 kJ/h, high fat meal 27.8 kJ/h [20]; high CHO meal 57.8 (19.1) kJ/h and moderate fat meal 49.8 (21.6) kJ/h [25]. No p values were provided for these comparisons; therefore, it is not known if these comparisons were statistically significantly different [20,25].
One study provided isocaloric comparisons and found no significant effect on DIT between high CHO, low fat meals and low CHO, high fat meals [31]. This study [31], which was described in Section 3.4 (higher energy vs. lower energy intake), provided the same group of subjects with high CHO, low fat meals of two different energy contents (2510 kJ and 5040 kJ), as well as low CHO, high fat meals of two different energy contents (2520 kJ and 5040 kJ) [31]. The DIT data were as follows: 5040 kJ high CHO, low fat meal 71.2 (15.5) kJ/h and 2520 kJ high CHO, low fat meal 45.6 (9.3) kJ/h vs. 5040 kJ low CHO, high fat meal 68 (12.4) kJ/h and 2520 kJ low CHO, high fat meal 45.6 (10.8) kJ/h, p > 0.05 for kJ/5 h comparing high CHO, low fat meals with low CHO, high fat meals [31].
Additionally, only one study [15] compared consuming isocaloric meals rich in protein vs. fat vs. CHO in participants of the same sex (females consumed 2500 kJ and males 3000 kJ). The high CHO and fat meals had the same DIT, whereas the high protein meal had a higher DIT (CHO meal 39.2 kJ/h, fat meal 39.2 kJ/h, and protein meal: 45.9 kJ/h, p < 0.01 for %/5 h ECM comparing four meals Model 2, adjusted for percentage of males, BMI, age, and hours of DIT measurement, also predicted a small but significant increase in DIT for every kJ intake (coefficient 0.012, standard error 0.0013, p < 0.001; CI: 0.0091; 0.014). This model predicts that for every 100 kJ increase in energy intake, DIT increases by 1.2 kJ/h. In this model, 16 studies were included with a total of 48 arms. Three studies included in model 1 were not included in model 2 because they had missing values for one or more of the variables investigated [16,25,34]. DIT accounted for 47.4% of the variance in Model 1 and 70.6% of the variance in Model 2. (an alcohol meal was excluded for the purpose of this SR)) [15].
One study [38] investigated the effect of consuming an adequate level of protein (3131 kJ) with a low level of protein (3114 kJ) in meals with similar energy contents. This study found a higher DIT when an adequate level of protein was consumed compared to a lower level (adequate protein meal 22.4 (5.7) kJ/h; low protein meal 7.8 (1.0) kJ/h, p = 0.001 for kJ/6 h and %/6 h) [38].
Riggs et al. [24] undertook isocaloric comparisons of meals differing in the amount of fat provided and found a higher DIT after consuming a moderate fat meal (1841 kJ) than an isocaloric low fat meal (1841 kJ), where both meals were high in protein, among normal weight participants (p < 0.005 for in %ECM) but not in overweight or underweight participants [24]. The DIT results were as follows; normal weight: moderate fat meal 43
A fixed model meta-analysis was conducted with these three studies [14,17] to compare DIT (kJ/h) for the MCT vs. LCT arms. Because Kasai et al. (2002) in study 2 [14] administered two interventions for MCT (margarine or mayonnaise) and two interventions for LCT (margarine or mayonnaise) to the same people, two forest plots are presented (one with only the margarine interventions and the other one with only the mayonnaise intervention arms). This avoids the effects of repetition of the same participants in both the margarine and mayonnaise studies. Both analyses found a significantly higher DIT when MCT was consumed compared to LCT (p = 0.002; Figure 3a,b). For both models the heterogeneity is 0% with chi 2 = 0.3 and p = 0.9 (Figure 3a), and chi 2 = 0.7 and p = 0.7 (Figure 3b). -study 2-mayonnaise trial-51.8% weight [14]. The % contribution of each study to the outcome is indicated as % weight.

Structure of Fats
Bendixen et al. [37] compared consuming meals with either a conventional fat (sunflower oil) or a chemically structured fat (rapeseed oil and octanoic acid by esterification with sodium methoxide) or a lipase-structured fat (rapeseed oil and octanoic acid by esterification with lipoxime IM) or a physically mixed fat (blending rapeseed oil and trioctanoate) [37]. The mean energy content of these four meals was 4698 (550.2) kJ [37]. This paper found a significant effect of fat structure on DIT with the highest DIT associated with the meal containing a chemically structured fat and the lowest with the meal having the conventional fat (conventional fat meal 61.8 (15.2) kJ/h, chemically structured fat meal 72.8 (19.0) kJ/h, lipase-structured fat meal 69.2 (11.4) kJ/h, and physically mixed fat meal 65 (13.9) kJ/h, p = 0.005 for kJ/5 h) [37].

Processed vs. Unprocessed Food
Only one study [26] compared consuming two meals with different levels of processing. One meal was composed of whole food (multi-grain bread and cheddar cheese either as one and a half sandwiches or two sandwiches) and the other was composed of processed foods (white bread and a -study 2-mayonnaise trial-51.8% weight [14]. The % contribution of each study to the outcome is indicated as % weight.

Structure of Fats
Bendixen et al. [37] compared consuming meals with either a conventional fat (sunflower oil) or a chemically structured fat (rapeseed oil and octanoic acid by esterification with sodium methoxide) or a lipase-structured fat (rapeseed oil and octanoic acid by esterification with lipoxime IM) or a physically mixed fat (blending rapeseed oil and trioctanoate) [37]. The mean energy content of these four meals was 4698 (550.2) kJ [37]. This paper found a significant effect of fat structure on DIT with the highest DIT associated with the meal containing a chemically structured fat and the lowest with the meal having the conventional fat (conventional fat meal 61.8 (15.2) kJ/h, chemically structured fat meal 72.8 (19.0) kJ/h, lipase-structured fat meal 69.2 (11.4) kJ/h, and physically mixed fat meal 65 (13.9) kJ/h, p = 0.005 for kJ/5 h) [37].

Processed vs. Unprocessed Food
Only one study [26] compared consuming two meals with different levels of processing. One meal was composed of whole food (multi-grain bread and cheddar cheese either as one and a half sandwiches or two sandwiches) and the other was composed of processed foods (white bread and a processed cheese either as one and a half sandwiches or two sandwiches) [26]. Subjects could choose to consume either one and a half sandwiches (2520 kJ) or two sandwiches (3360 kJ), and this choice was kept constant for both trials, thus the two trials were isocaloric for the same participant. There was a highly significant increase in DIT after consuming the whole food meal compared to the more processed meal (whole food meal: 99.4 (40.7) kJ/h; processed meal: 63.9 (35.6) kJ/h, p < 0.001 for total kJ and p < 0.01 for total % ECM [26]).

One Bolus Event vs. Isocaloric Smaller Frequent Meals
Four studies [27,32,34,35] compared administering a meal as a bolus event versus splitting the same meal into two [32], three [34], four [35] or six [27] smaller equal meals or snacks to be consumed throughout the morning. The time between multiple meals was 180 min [32], 60 min [35], or 30 min [27,34]. Kinabo and Durbin [32] compared two eating patterns using two different meal compositions: high CHO, low fat and low CHO, high fat. All four studies had the same participants perform both interventions (total n = 55). The energy density was as follows: 5040 kJ or 2510 kJ × 2 either as high CHO and low fat meal or low CHO and high fat meal [32]; 3150 or 1050 kJ × 3 [34]; 2823.4 kJ or 705.8 KJ × 4 meals [35]; and 3138 kJ or 523 kJ × 6 [27].
Two studies [32,34]  In order to clarify these discrepancies, a meta-analysis of the mean differences between bolus and smaller frequent meal event trials with fixed effects was conducted in RevMan to find the overall effect on DIT [32]. For these analyses, DIT was compared in kJ/h in order to standardize the units between studies. The forest plot shows ( Figure 4) the mean of the difference between bolus and smaller frequent meal event trials for each study. The overall mean of the difference is positive, which means that the DIT was lower in the smaller frequent meals event trials compared to the bolus trial. This overall effect on DIT was significant (p = 0.02). The heterogeneity was 14%, chi 2 = 4.6 and p = 0.3.
processed cheese either as one and a half sandwiches or two sandwiches) [26]. Subjects could choose to consume either one and a half sandwiches (2520 kJ) or two sandwiches (3360 kJ), and this choice was kept constant for both trials, thus the two trials were isocaloric for the same participant. There was a highly significant increase in DIT after consuming the whole food meal compared to the more processed meal (whole food meal: 99.4 (40.7) kJ/h; processed meal: 63.9 (35.6) kJ/h, p < 0.001 for total kJ and p < 0.01 for total % ECM [26]).

One Bolus Event vs. Isocaloric Smaller Frequent Meals
Four studies [27,32,34,35] compared administering a meal as a bolus event versus splitting the same meal into two [32], three [34], four [35] or six [27] smaller equal meals or snacks to be consumed throughout the morning. The time between multiple meals was 180 min [32], 60 min [35], or 30 min [27,34]. Kinabo and Durbin [32] compared two eating patterns using two different meal compositions: high CHO, low fat and low CHO, high fat. All four studies had the same participants perform both interventions (total n = 55). The energy density was as follows: 5040 kJ or 2510 kJ × 2 either as high CHO and low fat meal or low CHO and high fat meal [32]; 3150 or 1050 kJ × 3 [34]; 2823.4 kJ or 705.8 KJ × 4 meals [35]; and 3138 kJ or 523 kJ × 6 [27].
Two studies [32,34]  In order to clarify these discrepancies, a meta-analysis of the mean differences between bolus and smaller frequent meal event trials with fixed effects was conducted in RevMan to find the overall effect on DIT [32]. For these analyses, DIT was compared in kJ/h in order to standardize the units between studies. The forest plot shows ( Figure 4) the mean of the difference between bolus and smaller frequent meal event trials for each study. The overall mean of the difference is positive, which means that the DIT was lower in the smaller frequent meals event trials compared to the bolus trial. This overall effect on DIT was significant (p = 0.02). The heterogeneity was 14%, chi 2 = 4.6 and p = 0.3.  weight. Weight refers to amount of influence that the study exerts on the meta-analyses. The % contribution of each study to the outcome is indicated as % weight.      There was not a significant difference in DIT between palatable and unpalatable meals.
Data are described in mean (SD) unless otherwise described. a, b, c, etc. = these letters describe the types of meal interventions provided as illustrated in Table 1. † These data (mean and/or SD and/or 95% CI) were calculated or converted for one or more of these possible calculations or conversions (DIT % ECM calculated from DIT kJ, DIT % above baseline RMR calculated from DIT kJ, DIT kJ calculated from DIT % ECM, macronutrient % ECM calculated from grams, kcal converted to kJ, MJ converted to kJ, SE converted to SD, DIT % ECM or KJ or % above baseline RMR converted for unit of time, formulas described either in methodology or Supplementary Materials Table S1). † † Kinabo et al.

Fast vs. Slow/Normal Meal Consumption
Two studies [28,29] compared consuming the same isocaloric meal quickly or more slowly. One study [28] compared eating a meal (1255.2 kJ) as fast as possible to a meal chewed as many times as possible until no lumps remained before swallowing. The other study [29] compared eating a meal (1464 kJ) in 15 min compared to 5 min. Both studies found a higher DIT when the meal was consumed by slower eating compared to fast eating (slower eating 502.1 (234.4) kJ/kg/h vs. fast eating 19.5 (142.2) kJ/kg/h, p < 0.05 for kcal/kg/90 min [28]; slower eating 41.9 (14.6) kJ/kg/h vs. fast eating 31.6 (15) kJ/kg/h, p > 0.05 for cal/kg/180 min) [29]), although only one of the studies reached statistical significance [28].

Palatable vs. Unpalatable
Two studies [16,19] compared consuming palatable vs. unpalatable isocaloric meals (2930 kJ [19]; 2000 kJ [16]) on DIT. There was no significant difference in DIT between these two approaches indicating palatability did not influence DIT. In the first study [19], the effects of palatability were examined in both young and old participants (old participants: palatable meal 37

Discussion
This review investigated the effects of meals consumed after an overnight fast that differed in energy content or macronutrient composition on DIT, as well as the effects of consuming the same meal as a single event or multiple small meals or snacks. Studies comparing the effects of differing energy intakes supported a conclusion that a higher energy intake resulted in a higher DIT. This finding was further supported by two meta-regressions (one unadjusted and one adjusted for confounding factors), which found that for every 100 kJ increase in energy intake, DIT increased by 1.1 (unadjusted) or 1.2 (adjusted) kJ/h. A number of studies compared the effects of meals differing in macronutrient composition. One study found that a meal high in protein resulted in a higher DIT than meals high in CHO or fat, and a number of studies suggested that a meal high in CHO resulted in a higher DIT than a meal high in fat. Medium chain triglyceride meals produced a higher DIT than long chain triglycerides, the effects of mono-and polyunsaturated fats compared to saturated fats were unclear, fat structure (e.g., sunflower oil compared to a chemically structured fat) influenced DIT, and the fat content of a meal had inconsistent effects on DIT. The DIT of meals consumed as two or three small meals did not differ to the DIT of the same meal consumed as a single meal, whereas meals consumed as four to six small meals had a lower DIT compared to the same meals consumed as a single meal. Together these findings indicate that meals consumed after an overnight fast result in a DIT and the magnitude of this DIT is influenced by the energy content, the macronutrient composition, and the eating pattern of the meal.
Five studies investigated the effects of consuming different energy intakes on DIT as a primary outcome [21][22][23]31,36]. The study with the largest sample size found a significant increase in DIT with a higher energy intake [31]. Two studies with much smaller sample sizes reported trends of a higher DIT with a higher energy intake [23,36]. Two other studies [21,22] found no effect on DIT but the small sample sizes (eight and 11) could have impacted these findings. Additionally, one of these two studies provided little difference in energy intake between the two meals consumed [22]. The meta-regressions subsequently undertaken to examine the effect of energy intake on DIT across a much larger range of studies clearly support a conclusion that the energy content of meals consumed after an overnight fast influences DIT. Both the unadjusted meta-regression and the one adjusted for four confounding factors (percentage of males, age, BMI, and duration of DIT measurement) found similar significant relationships between a higher energy intake and a higher DIT. The magnitude of the increase in DIT was very small (1.1 or 1.2 kJ/h increase with each 100 kJ increase in energy intake), and whether this increase is clinically meaningful may depend on the magnitude of the energy content of the meal. These findings are consistent with the conclusion of Westerterp that energy intake is a predictor of DIT [41].
A number of studies compared the effects of meals consumed after an overnight fast differing in macronutrient composition. Five studies compared high fat vs. high CHO meals, and four of them found a higher thermogenic effect after the consumption of a high CHO meal compared to a high fat or moderate fat meal. The two studies that found significant effects had sample sizes of 24 males [33] and 14 males [30]. The other two studies, which were conducted with smaller sample sizes (12 males [25] and six females [20]), showed trends for a higher thermogenic effect of high CHO meals, but they did not provide p values and this limited their conclusions [20,25]. Furthermore, Thyfault et al. [25] compared a high CHO meal with a moderate fat, moderate CHO meal (45% fat) and therefore, the moderate CHO content could have confounded the findings. The one study that reported no significant difference for this high fat vs. high CHO comparison was conducted in 16 females [31]. Significant effects were found in the two studies conducted in males, whereas no significant effect was observed in the study in females, suggesting that males and females may respond differently following the consumption of CHO and fat meals. The differences in DIT between males and females may result from hormonal and/or body composition differences. Again, more research is needed to clarify these observations.
Two studies investigated the effects of protein on DIT; one found a significant increase in DIT after a high protein meal compared to high CHO or high fat meals in 19 participants of mixed gender [15]. The other study found a significant increase in DIT when a high protein meal was consumed compared to an adequate protein meal even though this was a small study of only six females [38]. Both studies suggest a high thermogenic effect of protein, however due to the limited numbers of studies, more research is needed to further investigate the thermogenic effect of protein in meals in both males and females, as gender was suggested to have an effect on the studies comparing high CHO vs. high fat meals [30,31,33].
A review by Tappy et al. [42] supports the higher thermogenic effect of protein compared to fat; this review reported DIT to be 0%-3% for fat, 5%-10% for CHO, and 20%-30% for proteins [42]. The different thermogenic effects of macronutrients are further reinforced by two studies comparing two diet interventions [43,44]. One study compared a high protein diet to an adequate protein diet for four days in 12 women and found a significantly higher DIT with the high protein diet (high protein diet 0.91 (0.25) MJ/day or 10.1 (2.7) % energy intake vs. high fat diet 0.69 (0.24) MJ/day or 7.6 (2.5) % energy intake, p < 0.05) [43]. Another study compared a high protein diet to a high fat diet for 36 h in eight women and found a significantly higher 24 h DIT with the high protein diet intervention as opposed to the high fat diet (high protein diet 1295 (240) kJ/day or 14.6 (2.9) % energy intake vs. high fat diet 931 (315) kJ/day or 10.5 (3.8) % energy intake, p = 0.02) [44]. Therefore, the findings of this SR regarding the role of meals differing in fat, CHO and protein composition on DIT are consistent with other studies investigating the effect of diets varying on macronutrients compositions on DIT. With regard to possible mechanisms of action of the higher thermogenic effect of protein, Westerterp-Plantenga [45] suggests that a higher protein diet may increase protein synthesis, which has a high energy cost, or if protein is oxidized the energy cost is higher than fat or CHO, and energy cost also varies with amino acid composition [45,46].
One study compared higher vs. lower levels of fat intake and found a significant increase in DIT following the consumption of a high fat meal compared to a low fat meal in female participants. This significant effect was only found in the 12 normal weight participants [24] and not in the six overweight or three underweight participants, however the small numbers of participants in the overweight and underweight groups would make finding effects in these groups difficult. Also, the different protein levels of the two meals could have impacted the findings of this study.
A number of studies compared the effects of different types of fats on DIT whereas no studies were identified that compared the effects of different types of proteins or carbohydrates on DIT. A significantly higher thermogenic effect was found when meals containing MCTs were compared to those containing LCTs; this finding was consistent in the three studies included [14,17]. The meta-analysis combining these three papers confirmed this significant increase in thermogenesis following the consumption of a meal with MCTs compared to LCTs (p < 0.005). Two hypotheses have been proposed regarding possible mechanisms by which the higher thermogenic effect of MCTs vs. LCTs might be achieved. One suggests an important role for the liver. MCTs are transported directly to the liver by the portal vein whereas LCTs are transported by the lymphatic system to peripheral tissues (adipose tissue and muscle) [47]. Also, LCTs need to bind to carnitine in order to pass through the mitochondrial membrane of the liver where B-oxidation occurs [47], whereas MCTs do not [48]. Therefore, MCTs being directly transported to the liver and that are easily able to pass through the mitochondrial membrane may be responsible for their higher DIT [49,50].The second hypothesis suggests a role for the sympathetic nervous system. Dullo et al. [51] found increased noradrenaline levels after MCT consumption and the authors suggested that sympathetic nervous system stimulation could therefore be responsible for the increase in energy expenditure of MCTs. Kasai et al. [14] has indicated that more research is needed to support this proposed mechanism.
Inconsistent results were found between two studies that compared different saturation of fat on DIT. The study which found a significantly higher DIT by PUFA and MUFA meals compared to a SFA meal [39] had a much larger sample size (29 participants) than the study which did not find a significant difference between MUFA and SFA meals (14 participants) [18]. More research is needed to clarify these findings. The effects of fat structure were also investigated. One study [37] found that fat structure, specifically meals containing a chemically structured fat, a lipase-structured fat, and physically mixed fat was associated with a significantly higher DIT compared to a conventional type of fat, and the highest DIT was associated with the consumption of the chemically structured fat. These findings are clearly preliminary and more studies are needed to confirm these observations.
One study examined the thermogenic effect of a less processed (e.g., whole grain bread) meal compared to a more processed meal [26] after an overnight fast and found a significantly higher DIT following consumption of the less processed meal. Although energy intake was consistent between trials in the same participant, the sample consisted of males and females, and there were two different caloric options within the study. It is unclear whether the choice of caloric options was controlled for in the analysis or whether there was a gender difference in the choice of caloric option. Furthermore, the macronutrient composition of these two meals differed and could have impacted the findings. Therefore, there is a need for further research comparing the effects of consuming more processed vs. less processed foods with the macronutrient content of the meals closely matched.
Four studies investigated the effects of consuming the same amount of calories and meal composition as a bolus event compared to a number of smaller meals during the morning. Two studies found a significant increase in DIT when the meal was consumed as one meal (bolus) instead of four [35] or six [27] smaller frequent meals. Two other studies did not find a significant difference in DIT between bolus and two [32] or three [34] smaller frequent meals regardless of macronutrient compositions [32,34]. The studies that did not find a significant difference provided less frequent meals for the snacking comparison, resulting in fewer meals with higher energy intake. The meta-analysis conducted on these four studies found that DIT was significantly higher when the meal was consumed as one bolus event. Together these results suggest that fewer larger meals result in a higher DIT than more frequent smaller meals.
Two studies [28,29] found a higher DIT following a meal eaten slowly compared to a meal eaten quickly, although these findings were only significant for one study [28]. The study that found a significant effect was conducted in ten males whereas the study that did not find a significant difference was conducted in nine females [29]. It is possible that gender is a factor influencing these results. Furthermore, these two studies suggest that the time that is spent on chewing the food may influence the magnitude of DIT; however more research is needed to clarify this observation and to compare the effects between males and females. Only two studies [16,19] have compared the effects of consuming a palatable versus an unpalatable meal on DIT in 19 males [19] or 12 participants (6 males and 6 females) [16]. Neither study found any significant differences in DIT. Although these findings suggest that palatability has no effect on DIT, the small number of studies limits the ability to draw any firm conclusions on this topic. Finally, it is important to note that no studies identified for this review have investigated the effects of differing micronutrients on DIT and this may potentially be another factor to influence DIT, which therefore warrants further investigation.

Strengths of This SR
This SR, including meta-analysis and meta-regression, is the first one to be conducted to investigate the effects of energy intake, macronutrient composition and eating events on DIT. This SR was able to identify and summarize the highest level of evidence available in this area and highlights where more research is needed in this field. The studies included were screened for quality/risk of bias and none of the studies had negative quality. Furthermore, the meta-regression was able to quantify the short-term effects of differing energy intakes after an overnight fast on DIT. In addition, the meta-analyses were able to quantify the influence of MCT vs. LCT and the role of consuming bolus eating events vs. smaller frequent meals in the morning on DIT. Considering the lack of evidence base regarding the role of meals consumed after an overnight fast on obesity prevention, this SR was able to provide evidence of the short-term effect of consuming different types of meals on DIT, which in the long term could play a significant role in obesity.

Limitations
There are a number of limitations regarding this SR. First, the included studies were very heterogeneous, differing in their research questions and the types of meals served after an overnight fast (as summarized in Tables 1 and 2). This heterogeneity limited the meta-analyses that could be conducted and the confidence with which conclusions could be drawn. Furthermore, only meta-analyses with a minimum of three studies were included in this SR.
Secondly, the majority of studies investigating DIT are conducted after an overnight fast even if their primary aim is not to investigate the effect of breakfast per se. These studies met the criteria to be included because they administered meals/snacks (even if not a typical breakfast meal) after an overnight fast (definition of breakfast). Lacking are studies of the effects of meals representative of breakfast in specific cultures. Whether these representative breakfast meals would have a different effect on DIT is unclear.
Thirdly, the studies varied in the units of measurement used to report DIT, including kcal or kJ, % ECM, or % above baseline (AB), and this limited the direct comparability of the findings between the studies. In order to address this issue, whenever possible, DIT was converted into units of measurements that allowed direct comparisons (e.g., DIT kJ converted to % ECM or % AB). Also, the studies measured DIT for different lengths of time, ranging from one and a half to six hours, and the length of time that the DIT is measured affects the magnitude of DIT detected [12,13,52]. In order to account for this limitation, the data were transformed into kJ/h allowing the results to be compared among studies. Furthermore, this confounding factor was adjusted for in the meta-regression model.
There is conflicting evidence about the length of time that DIT needs to be measured to provide accurate results. Two papers have recommended measuring DIT for at least three hours. One conducted a study in ten participants and concluded that 3 h DIT measurements are sufficient as 76% of DIT is obtained during this period [52]. The other paper [12] analyzed data from six studies with a total of 103 subjects and also recommended measuring DIT for three hours, as they found that the majority of the DIT was measured by three and a half hours when either high or low energy intakes (ranging from 1.3 MJ to 2.6 MJ) were consumed and in both men and women [12]. Another study [13] conducted in 131 participants recommended measuring DIT up to six hours, as 3 h measurements underestimated the DIT response by 40% [13]. Therefore, it is possible that studies included in this review that measured durations of DIT shorter than six hours may have underestimated DIT. In addition, DIT was measured differently in the studies, as some measured DIT for short durations at regular intervals (interval ranges also differed among studies) whereas other studies measured it continuously, which may have affected the magnitude of DIT detected. Specifically, Piers et al. [53] found a significantly (p < 0.01) lower DIT when it was measured at intervals compared to continuously in the same five subjects [53]. Most of the studies included in this review measured DIT at intervals and therefore there is a risk that the DIT was underestimated.
The small sample size of the majority of the studies included is another limitation of this SR. Sample sizes ranged from four to 29 participants, which limits the statistical power and capacity to find small but possibly important effects, as well as limiting the generalizability of the findings. This makes similarity of study designs that can be included in meta-analyses and meta-regressions even more important.
The interventions provided in the studies included in the meta-regression on energy intake had substantial variations; for example, they had different macronutrient compositions, and in some instances the meals were administered differently (such as meals consumed as a bolus vs. smaller frequent meals). Furthermore, the studies included in the meta-analyses were few, and even then included some heterogeneity. For example, the studies included in the MCT and LCT meta-analysis had meals with different levels of LCT or MCT. Also, for the bolus vs. smaller frequent meals meta-analysis, the energy intakes differed between the studies included. Furthermore, the smaller frequent meals arm differed in the number of smaller meals provided between the studies (two, three, four and six). All these factors were not possible to be adjusted for and could have impacted the findings of the meta-regressions and meta-analyses.
Furthermore, the meta-analyses were conducted by considering the two groups for comparison as two different groups of people, even though in reality it was the same group of people repeated in cross-over designs. For this reason, the analyses conducted by RevMan used a more conservative approach, because it is more difficult to find statistical significance if two comparison groups consist of different people. However, the heterogeneity between studies is also expected to be underestimated for the same reason. Therefore, for the two meta-analyses conducted in this SR (MCT vs. LCT and meal vs. snacking), the heterogeneities are expected to be larger than the number provided and the p values are also expected to be even more significant (smaller) that the ones provided (in both cases they are already highly significant). The CIs are also believed to be smaller than the ones provided.
Also, in the findings of this SR, meta-regression and meta-analyses were based primarily on studies rated neutral for quality, thereby limiting the confidence with which conclusions could be drawn.
Finally, the included studies were almost entirely short-term (single meal interventions; only one study investigated the effects of 2 weeks of the same meal daily prior to measurement following a single meal) [36], therefore, whether the effects observed reflect, at least in part, the novelty of unfamiliar meals consumed after an overnight fast on DIT is unclear.
It was not possible to draw any conclusions about the effect of a routine breakfast on DIT. Therefore larger, longer-term experimental studies are needed to draw conclusions about these topics. Specifically, there is the need to investigate the long-term effects of regular consumption of low energy intake vs. high energy intake meals and meals varying in macronutrient composition on DIT, and whether these factors ultimately affect total daily energy intake and DIT or the weight of participants. There is also a need to compare the effects of these influences on DIT between regular breakfast eaters and skippers, as this was found to be an important factor to consider by a previous trial conducted in this area [54].

Recommendations
Most of the studies included in this SR were rated neutral instead of positive quality because they had not provided enough or clear information about the selection of participants, recruitment, or inclusion and exclusion criteria. Therefore, it is recommended for future studies in this area to provide more information about recruitment, inclusion and exclusion criteria of participants, and any risk of biases in selection of the subjects. This SR found heterogeneity between studies regarding the length of DIT measured. Therefore, this SR has identified the need for more clarity on how long DIT should be measured in order to provide accurate results, and in order to achieve more homogenous study designs in this field.
This SR also highlights the heterogeneous ways DIT is reported between studies (kJ or kcal, % of ECM, or % above baseline RMR). It is recommended that future studies provide DIT in all of these three units of measure (kJ or Kcal, % ECM, and % above baseline RMR) in order to allow easier comparison with other studies conducted in this field and future meta-analyses.
It is also recommended for future studies to provide the data regarding baseline RMR and energy content provided by the meal, as this will allow a more comprehensive picture of the study design and results.
The majority of the studies had very small sample sizes. It is therefore recommended that future studies either increase the sample size in order to improve the statistical power of the studies, or provide evidence that the sample size used was adequate to detect an effect.

Conclusions
This systematic review has consolidated the current evidence regarding the effects of variations in energy intake, macronutrient composition, and the pattern of meals consumption after an overnight fast on DIT. It has also identified a substantial number of questions that remain to be answered, and the high level of uncertainty around many of the influences on the effects of meals on DIT. There is an enormous scope for future high quality studies in this field of research. Consensus on the duration of DIT measurement and larger sample sizes are just two ways in which research in this area could be improved. Comparisons of the effects of manipulations of meals consumed after an overnight fast on DIT in males and females, different age groups, and those who are healthy or have a range of obesity-related health conditions would also be informative.